DEGREE OF STRATA OF SINGULAR CUBIC SURFACES

TRANSACTIONS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 353, Number 1, Pages 95 115 S 0002-9947(00)02585-X Article electronically published on August 9, 2000 DEGREE OF STRATA OF SINGULAR CUBIC SURFACES RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO Abstract. We determine the degree of some strata of singular cubic surfaces in the projective space P 3. These strata are subvarieties of the P 19 parametrizing all cubic surfaces in P 3. It is known what their dimension is and that they are irreducible. In 1986, D. F. Coray and I. Vainsencher computed the degree of the 4 strata consisting on cubic surfaces with a double line. To work out the case of isolated singularities we relate the problem with (stationary) multiple-point theory. 1. Introduction The purpose of this work is to determine the degree of certain strata of singular cubic surfaces in P 3 C. In the nineteenth century L. Schläfli [26] and A. Cayley [6] classified the cubic surfaces in P 3 according to the type of their singularities. We consider this classification with the point of view given by J. W. Bruce and C. T. C. Wall [5] 100 years later, based on the modern theory of singularities [2], [3]. Let us suppose that the cubic surfaces are irreducible and have only isolated singularities. Fixed a singular point P of a cubic surface C, thereare4typesof surfaces, according to the rank of a quadratic form associated to each one of them: surfaces with a conic node, a binode, a uninode and a triple point at P. Inside each one of these classes, the different singularities presented by a surface C are determined by the possible coincidences of the 6 lines passing through the fixed singular point P and contained in C. Besides of these, there are 4 strata of irreducible surfaces with non isolated singularities (that are, in fact, ruled surfaces with a double line) and 9 of reducible cubic surfaces. The strata resulting from this classification is the closure of quasi-projective subvarieties of the P 19 parametrizing all cubic surfaces in P 3. Regarding geometric properties of strata, it is known what their dimension is [5] and that the strata are irreducible [4]. Our aim here is to determine their degree. First of all, the degree of strata parametrizing reducible surfaces is easy to compute [30]. In the case of the 4 big strata given by the rank of a quadratic form, the degree can be obtained at least in two ways [30], [28]. On the other hand, in 1986, D. F. Coray and I. Vainsencher calculated the degree of the 4 strata of ruled cubic surfaces having a double line [8]. Their work motivates ours. Received by the editors March 15, 1999. 2000 Mathematics Subject Classification. Primary 14N05, 14C17; Secondary 14C05, 14C15. Key words and phrases. Enumerative geometry, strata of singular cubic surfaces, stationary multiple-point theory. 95 c 2000 American Mathematical Society

96 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO In this paper, we determine the degree of 7 of the 11 strata of cubic surfaces with a conic node, using a method that must work, essentially in the same way, with some strata of cubics with a binode or a uninode. Our approach to the problem is to relate it with stationary multiple-point theory, whose main goal is the enumeration of k-tuples of points of a scheme Z with the same image under a morphism of schemes f : Z S and such that the points have collapsed in accordance with a given partition of k. We construct first some suitable parameter spaces and then we consider the lines whose coincidences determine the strata in our case, as stationary multiple points of a morphism between two of these parameter spaces. One of the first examples of multiple-point formulas related with our problem is that of De Jonquières (1866), allowing us to consider coincidences of points in the zero scheme of sections of a line bundle on a given curve. One century later, I. Vainsencher [29] (1981) proved a generalized De Jonquières formula, valid for a general line bundle. Some years after, S. J. Colley [7] obtained more formulas for stationary multiple points using the iteration method by S. L. Kleiman [16], [17], [18], who had already given some multiple-point formulas for the non-stationary cases. We cannot apply Vainsencher s results in order to compute the strata s degree because we are dealing with coincidences of points in the plane. On the other hand, in Colley s work there are no formulas for the cases we need (stationary 6-tuple points). We use then the more general multiple-point formula by Z. Ran [24] (1984), based on the residual intersection formulas of W. Fulton and D. Laksov [10], that allows us to treat coincidences of points in the zero scheme, let s say X, of sections of a vector bundle with rank equal to codim(x). This turns out to be our case with a rank 2 vector bundle while in [24] all the examples worked out deal with line bundles. Then, in 5 we prove results justifying the computation of the degree of strata of cubic surfaces with a conic node, as a product of cycles in the intersection ring of a suitable space of 6-tuples. The cases not covered here are the strata given by a collapse with less than 3 different lines. Those correspond in our construction to 6-tuples of points that are not curvilinear (i.e. that don t live in a smooth curve) and, in general, the parameter space Z k used to enumerate them only parametrizes ordered and curvilinear k- tuples of a scheme Z over another one S [17], [24]. Regarding to the effective computation of these degrees, the method employed here leads to calculate in A (Z k ) (the intersection ring of Z k ) the degree of a zerodimensional cycle, obtained by taking up cycles in A (Z l ), with l<k, through the inductive construction of Z k. Given a morphism from Z to S, if A (S) isknownanda (Z) is a finitely generated A (S)-module, we are able to give an inductive method to compute the degree of zero-dimensional cycles of A (Z k ) in the previous conditions (see Proposition 6.1). We note that, whenever Z is a projective bundle over S, we can explicitly describe the intersection ring of the space Z k (see 6). In our case A (Z 6 )isapolynomial ring over Z with 23 generators modulo 38 relations and effective computations in this ring, using Gröbner basis method, is much more costly that our inductive method.

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 97 We implement this method in a computer program written in Maple [20]. This program could be useful each time that one uses the parameter spaces Z k in order to study enumerative questions depending on what is happenning on the fibers of Z over S. This paper is organized as follows: 2 is devoted to the classification of singular cubic surfaces in P 3. In 3, we construct some parameter spaces appropriated for the problem. We deal in 4 with the spaces of k-tuples and the secant bundles of Ran s work [24]. In 5 we state and prove our main result giving the degree of some strata of cubic surfaces with a conic node. Finally, in 6 weworkwith the intersection ring of spaces of k-tuples and the effective computation of strata s degree. This work is part of the second author s Ph.D. Thesis. We would like to thank E. Arrondo and I. Vainsencher for suggestions and comments. Notation 1.1. We shall tacitly assume that all schemes are defined over an algebraically closed field of characteristic 0. Our notation follows the book of W. Fulton [9], in particular we will write P (E) for the projectivization of a vector bundle E over a scheme Z, defined as P (E) =Proj(SymE ). If Z is a non-singular scheme, the intersection ring we shall use is the Chow ring of classes of rational equivalence of cycles, graduated by codimension and denoted by A (Z). We will denote by Gr the grassmannian of lines of P 3. 2. Classification of singular cubic surfaces The classification of singular cubic surfaces in P 3 givenbyj.w.bruceand C. T. C. Wall [5] corresponds to another one in accordance with the incidence relations of the lines contained in a cubic surface, studied by several authors [6], [25]. A cubic surface is non-singular if and only if it contains 27 different lines. When it is singular, some of the lines collapse. The surface given by a general point of each stratum presents a certain type of coincidence between these lines. We recall here the main points of Bruce s and Wall s work. If P is a singular point of a cubic surface C and (x 0 : x 1 : x 2 : x 3 ) are homogeneous coordinates in P 3, one can assume, doing a change of coordinates if necessary, that P is the point (0 : 0 : 0 : 1) and then, the equation of the surface takes the form (2.1) x 3 f 2 (x 0,x 1,x 2 )+f 3 (x 0,x 1,x 2 )=0 where f i is homogeneous of degree i. The classification depends, first on the rank of the quadratic form f 2, and second, on the coincidences between the collection of 6 lines common to the two cones given by f 2 =0andf 3 = 0. These are lines through P, contained in C and they are equivalent to the intersection points of the two plane curves, let s say V 2 and V 3,givenbyf 2 (x 0,x 1,x 2 )=0andf 3 (x 0,x 1,x 2 )=0 respectively. Let us suppose that the surface is irreducible and has only isolated singularities. According to the rank of f 2 there are 4 types of singularity at the fixed point P : conic node, binode, uninode and triple point. When the rank of f 2 is maximum, the singular point is a conic node. There are 11 strata in accordance with the partition of 6 given by the coincidences of the intersection points (lines) of V 2 and V 3,namely

98 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO 1 6 21 4 31 3 2 2 1 2 41 2 321 2 3 Sing. A 1 2A 1 A 1 A 2 3A 1 A 1 A 3 2A 1 A 2 4A 1 Param. 3 2 1 1 0 0 0 51 42 33 6 Sing. A 1 A 4 2A 1 A 3 A 1 2A 2 A 1 A 5 Param. 0 0 0 0 In this table, the partition called, for example, 2 2 1 2 corresponds to a collection of 6 points with 2 of them being double points. At a singular point of type A k, the normal form of the surface s equation is x k+1 0 + 3 j=1 x2 j =0. Thethirdrow indicates the number of parameters on which each type of surface depends, up to projective transformations. A binode corresponds to rank(f 2 ) = 2, in this case, P is a singularity of type A k with k 2. There are 13 possible configurations of common points to the cubic and the line pair and 6 of them correspond to new strata. 1 3.1 3 1 3.21 1 3.3 21.21 21.3 3.3 1 2.1 2 Sing. A 2 A 2 A 1 2A 2 A 2 2A 1 2A 2 A 1 3A 2 A 3 Param. 2 1 1 0 0 0 1 1 2.2 2.2 1 2.1 2.1 1 2.0 2.0 Sing. A 3 A 1 A 3 2A 1 A 4 A 4 A 1 A 5 A 5 A 1 Param. 0 0 0 0 0 0 When rank(f 2 )=1,P is a uninode and one gets three more strata called D 4, D 5 and E 6 corresponding to the cases in which the points in V 2 V 3 are three distinct points, two and one, respectively. Finally, in the case of f 2 =0,P is a triple point. It must happen that f 3 =0 defines a non-singular cubic in order to have an isolated singularity at P. The surface is then a cone over this plane cubic. Until now the singular point P is fixed on the cubic surface. When P moves in P 3, one gets strata in P 19 P 3 of cubic surfaces with a distinguished singular point. Its projection onto the first factor gives the strata of singular cubic surfaces, living in P 19, that we are interested in. 3. Parameter spaces We first construct a subscheme of P 19 P 3, that we will call S, parametrizing cubic surfaces in P 3 with a distinguished singular point. In this way, strata of cubic surfaces with a singular point can be seen as subschemes of S and the degree of corresponding strata of singular cubic surfaces in P 19, obtained by projection onto the first factor, could be computed, using the projection formula (see 3.2.(c) of [9]), as a product in the Chow ring A (S). 3.1. Cubics with a distinguished singular point. The construction of the parameter space S for cubic surfaces with a distinguished singular point is analogous to others appearing in the literature for plane curves (see, for instance, [22] and [23]). To fix notation, let V be a vector space of dim 4 over C and P 3 = P (V )the projective space of lines of V.Weset (3.1) 0 L V P 3 Q 0

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 99 for the tautological exact sequence with rank(l) =1andrank(Q) =3andwe denote by h = c 1 (L ) the first Chern class of L = O P 3(1) (h is the hyperplane class of P 3 )andbys 3 V the third symmetric power of V. Definition 3.1. Let F be the subbundle of S 3 V P 3 whose fiber over a point of P 3 is the linear subspace of S 3 V of cubic forms that have multiplicity at least 2 at such point. Denote by S the projective bundle associated to F. With this definition S = P (F ) P (S 3 V P 3) = P 19 P 3 is the incidence subvariety whose points consist in couples (f,p), where P is at least a double point for the cubic surface given by f =0. Sincerank(F ) = 16, then dim(s) = 18. In order to determine the intersection ring of S = P (F ), we start with a preliminary lemma. Lemma 3.2. With the previous notation, the sequence of vector bundles over P 3 (3.2) 0 S 2 Q Q u S 2 Q V P 3 S 3 Q v F 0 where u(α β) =(α β, αβ) and v(α ω, β) =αω + β, is exact. Therefore c(f )=1 8h +40h 2 160h 3. Proof. The first part is similar to the analogous result for plane curves (see [22]). The second one is a straightforward computation applying Whitney s formula (see 3.2.(e) of [9]) to this resolution and to the tautological exact sequence over P 3 above (3.1). Proposition 3.3. Let π be the projection from S to P 3. The intersection ring of S = P (F ) is A (P (F )) = Z[b, µ] (b 4,µ 16 8bµ 15 +40b 2 µ 14 160b 3 µ 13 ) where b = π h and µ = c 1 (O P (F ) (1)). Moreover, the following relations hold: b 3 µ 15 =1, b 2 µ 16 =8, bµ 17 =24and µ 18 =32. Proof. The description of A (S) as a polynomial ring is a direct consequence of the result giving the Chow ring of a projective bundle P (E) in terms of the total Chern class of E and the intersection ring of the base space (see 3.3.(b) and Ex. 8.3.4. of [9]). In order to verify the relations in A (S) we take down the products in this ring to A (P 3 ), using the projection formula and standard results about Chern classes (see Ex. 3.2.1. of [9]). 3.2. More parameter spaces. We build now a space, say X, that fixes a singular point P for each cubic surface C, and parametrizes the lines contained in C and passing through P. The coincidences between these lines determine the stratum where the surface lives. To get X, we first construct another space Z in P 19 P 3 Gr. This Z turns out to be a projective bundle over S and we will define X as a subscheme of Z. If we write f for the restriction to X of the projection from Z to S, it is natural to consider the stationary multiple points of f to solve our problem concerning the strata of singular cubic surfaces, because the condition defining the strata of cubics is the collapse produced between the points in the fiber of X over S.

100 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO Remark 3.4. One has to take care of three pairs of strata such that both strata on each pair have the same codimension in the P 19 of cubic surfaces and they correspond to the same type of coincidence between points of X over S. These pairs are (4A 1,D 4 ), (2A 1 A 3,D 5 )and(a 1 A 5,E 6 ). In this paper, we don t compute the degree of the strata of surfaces having singularities of the types appearing on the two last pairs but, in order to compute the degree of surfaces with four conic nodes (4A 1 ) one must substract the contribution given by surfaces with a uninode (D 4 ) (see Theorem 5.1). 3.2.1. Construction of Z. We consider the incidence of points and lines of P 3, I P 3 Gr, consisting of pairs (P, l), P L, wherel is the point of Gr corresponding to the line L P 3. Definition 3.5. If p 1 (resp. p 2 ) is the projection from I to P 3 (resp. Gr), Z is defined as the scheme I P 3 S (see B.2.3. of [9]). Z π I p 2 Gr. p 1 p 1 S π P 3. This Z represents collections (f C,P,l)withf C = 0 the equation of the cubic surface C, P a double point of C and L a line through P. By definition, Z is a projective bundle over S: the projection p 1 identifies the incidence I with P (T P 3), (T P 3 is the tangent bundle of P 3 ), so we have also (see B.5.5. of [9]) I = P (T P 3( 2)) = P (Q( 1)) and therefore Z = P (π Q( 1)) and dim(z) = 20. It is also easy to see that p 2O Gr (1) = O P(Q(-1)) (1), p 1O P 3 (1) = O P (U) (1) and p 2 : I Gr can be identified with the projective bundle P (U) overgr, beingu the tautological subbundle of rank 2 fitting in the tautological exact sequence over Gr: (3.3) 0 U V Gr Q Gr 0. These facts will be useful on later computations (see 3.2.2). 3.2.2. Construction of X. If C is a cubic surface in P 3 givenasthezeroesofa section f C of O P 3 (3), the Fano scheme of lines contained in C, F (C), can be seen as the scheme of zeroes of the section induced by f C in p 2 (p 1 O P 3 (3)) = S3 U. The line bundle p 1O P 3 (3) over I corresponds to the divisor (p 1f C ) 0 with fiber at a general l Gr the three intersection points of the cubic surface C with the line L of P 3. This is a particular case of the Fano scheme of pairs (V,l) wherev is a hypersurface in P n of given degree d and l parametrizes a line L contained in V, defined in [1]. There it is proved that, if H is the projective space of all hypersurfaces of degree d, G is the grassmannian of lines of P n and U is the tautological subbundle over G, then the Fano scheme in H G is given by the scheme of zeroes of the section of S d U O H (1), induced by the section of O P n (d) O H (1) defining the universal divisor of H P n (see Thm. 3.3. of [1]). To obtain a space X Z consisting on the points (f C,P,l)ofZ such that L C, we adapt this result to our case for which lines must be contained in the surface C

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 101 and must pass through the distinguished singular point P C (see Proposition 3.9 below). X Z P 19 P 3 Gr S P 19 P 3. First, we define a space T over Z which plays the role of the incidence I as projective bundle over Gr in the case of a concrete cubic surface: we want the fiber of T over a point (f C,P,l) parametrized by Z to represent the P 1 of points of the line L. We then consider the diagram (3.4) T π Y pr 2 I pr 1 pr1 p 2 Z p 1 S π π I p1 P 3 p 2 Gr p 1 P 3 where p 2 is the projection of P (U) Gr, Y := I Gr I = P (p 2 U) P3 Gr P 3 and T := Y I Z = P (π p 2U) P 19 P 3 Gr P 3. In this way, Y parametrizes collections (P, l, P ) such that P L and P L and T collections (f C,P,l,P )wherep is a double point of the surface C and P, P L. Now, we build an invertible sheaf N over T which is the analogue of p 1O P 3(3) in the case of F (C). Lemma 3.6. Let N bethelinebundlepr 1 p 1 O S (1) O T (3) over T.ThenN has an associated section s N such that D := (s N ) 0 parametrizes collections (f C,P,l,P ) with P C, besides P being a double point of C and P, P L. Proof. Let π 1 and π 2 be the projections from S P 3 to its factors. If φ is the morphism from T to S P 3 given by the diagram (3.4) then, by construction, N is the pullback by φ of π1 O S(1) π2 O P 3(3). The section s N on the statement above is φ of the section of π1 O S(1) π2 O P3(3) vanishing on the universal surface in S P 3. Remark 3.7. For each point z Z, the fiber of (s N ) 0 T consists of two points: P the double point of C and, let s say R, the remaining point in the intersection of a general line L through this P, and the surface C. Let D P be the divisor of T given by the point P when z moves in Z. Let D R be the residual divisor of D P on D (see 9.2.1. of [9]), given by the point R when z moves in Z. We will get X as the scheme of zeroes of a section of a rank 2-vector bundle over Z, defined as pr 1 of the line bundle over T corresponding to the divisor D R. Definition 3.8. Let E be the sheaf pr 1 (O T (D R )), where pr 1 is the projection from T to Z and D R is defined in the previous remark. Proposition 3.9. With the previous notation: 1. The pairs consisting of a flat family of cubic surfaces of P 3 withadistinguished singular point and a flat family of lines contained in the surface and passing through its singular point, both parametrized by the same scheme W are the W -points of a closed subscheme X of Z.

102 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO 2. The space X is the scheme of zeroes of a regular section, say s E,ofthelocally free sheaf E of rank 2 over Z. Proof. The proof is analogous to that of Thm. 3.3 (i), (iii) of [1] except for the following concerning 2: By definition, D =2D P + D R,soM := O T (D R )=N O T ( 2D P )andthe section s N of N induces another one of M, let ssays M, vanishing on D R. The sought section is equal to s E = pr 1 s M. E is a rank 2-vector bundle because by Grauert s theorem (see III.12.9 of [12]), E = R 0 pr 1 (O T (D R )) is locally free if dim H 0 (T z, O T (D R ) Tz ) is constant z Z and in our case, h 0 (T z, O T (D R ) Tz )=h 0 (P 1, O P 1(1)) = 2, z Z. We write j for the inclusion X Z, p for the projection from Z to S (called before p 1 )andf for the composition p j. The situation we get is E se X =(s E ) 0 j Z P 19 P 3 Gr p S P 19 P 3 f with the space X parametrizing, for each cubic surface with a given double point, the lines passing through that point and contained in the surface. In this way, the strata of singular cubic surfaces with a conic node of given type (see 2) depend on the coincidences between points on the fiber of f : X S, keeping in mind Remark 3.4. We will show in 5 that their degree can be determined using a suitable cycle of stationary multiple points of f and computing in the intersection ring of the space of 6-tuples Z 6 (see 4). In 5 we will need to know c(e), the total Chern class of the vector bundle E and also that X is smooth. Lemma 3.10. With the previous notation, if ξ = c 1 (O Z (1)) is the hyperplane class of Z and we denote by b, µ the pullback by p : Z S of the same classes in A (S), thene = π p 2U p 1 O S (1) π O P (U) ( 2) π O P (Q( 1)) (2) and therefore c(e) =1+(5ξ 4b +2µ)+(6ξ 2 +5µξ 9bξ + µ 2 4µb +3b 2 ). Proof. Since E := pr 1 (O T (D R )) and O T (D R )=O T (D) O T ( 2D P ), we first determine [D P ] A 1 (T ). If ξ T is the hyperplane class of T = P (π p 2U)) over Z, then [D P ]=a ξ T + pr 1 [D ] with D A 1 (Z) (see Ex. 8.3.4. of [9]). Now, by construction, [D P ] [T z ]=1and this implies a =1(forξ T [T z ]=1andthefibersofT over Z don t intersect each other). To get D, we compute pr 1 ([D P ] ξ T ) in two ways: 1. pr 1 ([D P ] ξ T )=pr 1 (ξt 2 )+[D ] pr 1 (ξ T )=π p 2 c 1(Q Gr )+[D ], where the first equality is the projection formula and the second is Ex. 3.2.1 of [9], being Q Gr the tautological quotient of rank 2 over the grassmannian. 2. Let s P be the section of T over Z such that Im(s P )=D P.Then pr 1 ([D P ] ξ T )=c 1 (s P O T (1)).

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 103 From1and2above,[D ]=c 1 (s P O T (1)) π c 1 (O P (Q( 1)) (1)) A 1 (T ), recalling that c 1 (Q Gr )=c 1 (O Gr (1)) and p 2 O Gr(1) = O P (Q( 1)) (1). To compute c 1 (s P O T (1)), we use the upper part of diagram (3.4): π pr 2 T Y I pr 1 sp tp pr1 p 2 Z π I p 2 Gr p 1 P 3 wherewedefinet P as the section of Y = I Gr I whose image is the diagonal of Y. Looking at the diagram we see that s P O T (1) = π t P O Y (1) (as T = P (π p 2U)). Now, by construction, t P c 1(O Y (1)) = c 1 (O I (1)) with I = P (U) overgr. Putting all these things together [D ]=π (c 1 (O I (1)) c 1 (O P (Q( 1)) (1))) and: - O T (D P )=O T (1) pr 1 π O P (U) (1) pr 1 π O P (Q( 1)) ( 1), - O T (D R )=O T (1) pr 1 (p 1 O S (1) π O P (U) ( 2) π O P (Q( 1)) (2)). Finally, by the projection formula: E = pr 1 (O T (D R )) = π p 2U p 1 O S (1) π O P (U) ( 2) π O P (Q( 1)) (2). From this expression, the computation of c(e) is straightforward. We see now that X is smooth (even though X is not smooth over S). Proposition 3.11. With the previous notation, X is smooth. Proof. We are going to verify that, if P P 3, then the fiber X P can be identified with the total space of a projective bundle over P 2 and therefore X is smooth. Since X =(s E ) 0 with s E a section of the vector bundle E over Z, thenx P Z P can be identified with the scheme of zeroes of the section of E ZP induced by s E. Now Z P = P 15 P 2 and, if τ 1 y τ 2 are the projections from P 15 P 2 to its factors, then E ZP = τ1 O P 15(1) τ 2 (O P 2(2) O P 2(3)). To see this we use the explicit description of E given in the previous lemma, noting that: The fiber I P consists on the plane in Gr of lines through the point P. p 2U IP = OP 2 O P 2(1). This can be checked recalling that U = p2 (p 1O P 3(1)), and so, every section ρ of U is induced by a section η of O P 3(1) in such a way that ρ vanishes on the plane Π Λ of Gr parametrizing the lines contained in the plane Λ which is the zero locus of η. In this way, the restriction of the Koszul exact sequence (see B.3.4. of [9]): 0 O Gr U I ΠΛ (1) 0, to I P is 0 O P 2 U P 2 O P 2(1) 0, for the intersection in Gr of a plane of lines through a general point of P 3 and a plane of lines contained in a general plane of P 3 is empty (see Ex. 14.7.2. of [9]). p 1 O S (1) ZP = OP 15(1) (recalling that S P = P 15 ).

104 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO The restriction of O P (U) ( 1) to I P is trivial since I = P (U) p2 Gr and O P (U) (1) = p 1O P 3(1). O P (Q( 1)) (2) IP = O P 2(2), for I = P (Q( 1)) p1 P 3. Now, if we call R := O P 2(2) O P 2(3), then P (H 0 (R)) = P 15 and the sequence of vector bundles over P 2 : 0 K H 0 (R) O P 2 ev R 0, given by the evaluation of sections of R, is exact. The key observation is that the tautological section Γ of τ1 O P 15(1) τ2 R = E ZP is precisely the restriction to the fiber Z P of s E. The situation is τ1 O P 15(1) τ 2 R Γ (Γ) 0 P (H 0 (R)) P 2 τ2 P (H 0 (R)) P 2 and so X P = (Γ)0 which is isomorphic to the projective bundle P (K) overp 2. Therefore X is smooth. 4. Spaces for k-tuples and secant bundles In 5, we shall use the spaces for k-tuples Z k and the secant bundles E k,forz and E defined in 3, to compute the degree of some strata of cubic surfaces with a conic node. We describe them here taking as reference the work by Z. Ran [24]. Spaces for k-tuples. The space Z k is built inductively as a suitable blow-up of an ordered k-product of Z. Steps 0, 1. One denotes Z 0 := S, Z 1 := Z, p 1 0 : Z 1 Z 0 is the morphism p and one defines ι j : Z j Z j as the identity i for j =0, 1. Step 2. Considering the fiber product defined by the cartesian diagram (see B.2.3. of [9]): pr 2 Z 1 Z0 Z 1 pr1 p Z 1 Z0, one defines a diagonal 1 =(i ι 1 )(Z 1 ) Z 1 Z0 Z 1 ; b 2, the blow-up of Z 1 Z0 Z 1 along 1 with exceptional divisor D1,2, 2 Z1 p D1,2 2 Z 2 b2 1 Z 1 Z0 Z 1 and new projections from Z 2 to Z 1 ; p 2 1,1 := pr 1 b 2 y p 2 2,2 := pr 2 b 2. Since p is smooth, 1 Z 1 Z0 Z 1 is a regular inclusion (see B.7.3. of [9]) and the new projections are also smooth. Out of the exceptional divisor, Z 2 parametrizes couples (z 1,z 2 ) of different points of Z having the same image in S, i.e., strict double points of f. InD 2 1,2 live the couples of points of Z having the same image by p that are infinitely near (b 2 (D 2 1,2) consists on pairs (z,z) wherez is a ramification point of p).

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 105 (4.1) One notes that (ι 1 pr 2, ι 1 pr 1 )isaninvolutionofz 1 Z0 Z 1 reversing the order in (z 1,z 2 ) and fixing the diagonal 1, so it lifts to an involution of Z 2 that will be denoted by ι 2. Induction hypothesis. One supposes that a space Z k of k-tuples and associated objects (D1,k k, pk 1,k 1, pk 2,k, ι k) are built as before. Step k + 1. One considers the fiber product defined by pr 2 Z k Zk 1 Z k Zk pr1 g ι k =p k 2,k Z k g=ι k 1 p k 1,k 1 Z k 1 where p k 1,k 1 := pr 1 b k, p k 2,k := pr 2 b k and b k : Z k Z k 1 Zk 2 Z k 1 is the previous blow-up. Next, one defines k =(i ι k )(Z k ), where ι k is (ι k 1 pr 2, ι k 1 pr 1 ) lifted to Z k (ι k is a regular inclusion by B.7.3. of [9]) and the blow-up along k with exceptional divisor D k+1 1,k+1, b k+1 : Z k+1 Z k Zk 1 Z k. In this way, the points of Z k correspond to ordered k-tuples of points of Z having the same image by p in S. By B.7.3. of [9], it holds N k Z k Zk 1 Z k = ι k T ιk 1 p k 1,k 1 = Tp k 2,k. This will be useful in 6, in order to make computations in the intersection ring A (Z k ). Remark 4.1. Note that in [24] g is taken as p k 1,k 1 and so k =(i ι k )(Z k )isnot contained in Z k Zk 1 Z k. Regarding the enumeration of stationary multiple points of the morphism p, given a partition σ =(k 1,k 2,..., k r )ofk with r i=1 k i = k, Ran defines a subscheme of Z k called V σ, parametrizing k-tuples of Z/S with the first k 1 points being infinitely near to each other (i.n.), the next k 2 points being also i.n., etc. For example, the subscheme of Z 4 parametrizing pairs of points of type (2, 2) is V (2,2) =(p 3 1,2 p4 1,3 ) D1,2 2 (p4 2,4 ) D1,3 3 for points in Z 4 correspond to collections (z 1,z 2,z 2,z 3,z 3,z 2,z 2,z 4 )ofz/s. Anyway, for each partition σ of k, one can define a subscheme of Z k parametrizing k-tuples of type σ as a suitable intersection of (pullbacks by projections of) exceptional divisors reflecting the coincidences given by σ. Remark 4.2. The space Z k had already been defined by S. L. Kleiman (see p. 390 of [16] and [17]). For the construction of Z k+1, Kleiman considers the fiber product Z k Zk 1 Z k containing the usual diagonal (i i)(z k ). For the justification of Z k as parameter spaces, see [18] and [24] where a relation between Z k and the relative Hilbert scheme Hilb k (Z/S) is established, showing that Z k is a parameter space for ordered and curvilinear k-tuples of points of Z with the same image in S (a collection of points is curvilinear if the points live in a smooth curve). Consequently, the (stationary) multiple points formulas given by Colley, Kleiman and Ran cannot enumerate the non-curvilinear points of a morphism. Secant bundles. The notion of secant bundle, introduced by R. L. E. Schwarzenberger [27] in 1964, generalizes the concept of tangent bundle of a projective variety Y in the following sense: the secant bundle of Y is defined over a product scheme of Y so that its restriction to the diagonal in this product is the tangent bundle of

106 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO Y. The Chern classes of secant bundles are related with multiple-point cycles (see [21], [19] and [29]). By definition, a secant bundle is always associated to a line bundle over a scheme. In [24], given a vector bundle E of arbitrary rank over a smooth scheme Z, a secant bundle E k over Z k is defined, allowing us to compute the class [X k ]ina (Z k ), when X is the zero subscheme of a section of E. The E k are defined inductively and they fit into an exact sequence of vector bundles over Z k (analogous to the one given by Schwarzenberger for its secant bundles) from which one can compute its Chern classes. Regarding the definition of E k : for k =1,E 1 := E; ifk =2,E 2 is defined by the exact sequence over Z 2 : 0 E 2 (λ 1 2,λ1 2 ) (p 2 1,1 ) E (p 2 2,2 ) E ρ2 (p 2 1,1 ) E D 2 1,2 0, with ρ 2 (a, b) =res(a) res(b), and res the restriction to the exceptional divisor D 2 1,2. E 2 is a vector bundle over Z 2 by Nakayama s lemma and from its definition it is easy to see that 0 (p 2 2,2 ) E L D 2 1,2 E 2 (p 2 1,1 ) E 0 is exact, where in general L D is the line bundle associated to the divisor D. When k 3, by induction, one supposes that, j k, vector bundles E j and applications are defined. Then { λ 1 j : E j (ι j 1 p j 1,j 1 ) E j 1 λ 2 j : E j (p j 2,j ) E j 1 Definition 4.3. Let J k+1 := (ι k p k+1 1,k ) E k (p k+1 2,k+1 ) E k and let K k+1 be the vector bundle given by the exact sequence over Z k+1 : }, (κ 0 K 1,κ 2 ) ρ k+1 k+1 J k+1 (p k+1 2,k+1 ) E k D k+1 1,k+1 with ρ k+1 (a, b) :=res(a) res(b). Let E k+1 be defined by 0, (λ 1 k+1 0 E,λ2 k+1 ) ν k+1 k+1 K k+1 (p k+1 2,k ) E k 1 L D k+1 0, 1,k+1 with ν k+1 = (((p k+1 1,k ) (λ 2 k ) (pk+1 2,k+1 ) (λ 1 k )), p k+1 2,k := ι k 1 p k 1,k 1 pk+1 2,k+1 : Z k+1 Z k 1. A standard computation using commutative diagrams shows that with this definition the sequence of vector bundles over Z k+1 : (4.2) 0 (p k+1 k+1,k+1 ) E L D k+1,k+1 E k+1 (p k+1 1,k ) E k 0 where p k+1 k+1,k+1 := p2 2,2 pk 2,k pk+1 2,k+1 and D k+1,k+1 := D k+1 1,k+1 +(pk+1 2,k+1 ) D k 1,k +... +(p 3 2,3 p 4 2,4 p 5 2,5 p k+1 2,k+1 ) D 2 1,2 is exact (for details see [30]).

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 107 Remark 4.4. Note that our definition of E k for k 3 is different from the one given in [24] for which, an easy computation, using Whitney s formula (3.2.(e) of [9]), shows that the total Chern class of E k+1 is not the one given by the exact sequence (4.2) with which, in some examples, one recovers known results (see [30]). 5. Degree of some strata of cubic surfaces with a conic node Let σ =(k 1,k 2,..., k r ) be a partition of 6 with k j k l,ifj l, C σ the stratum in P 19 of cubic surfaces with a conic node of type σ, C U the one of surfaces with a uninode and α σ := dim(c σ ). In principle, we would get deg(c σ )as#(c σ Λ) being Λ P 19 a general linear space of codimension α σ. The stratum C σ depends on coincidences between points on the fibers of the map f : X S, unless in the case of a coincidence of type σ =(2, 2, 2) that corresponds to the union of C (2,2,2) and C U (see Remark 3.4). Since X Z is not smooth over S, we consider the space Z 6 of 6-tuples of Z/S treated in 4. Even though f is not smooth, the inclusion j : X Z induces, by construction, inclusions j k : X k Z k. Let φ 1 denote the projection from Z 6 to Z assigning to each 6-tuple of Z 6 its first element and τ the projection from S to P 19.Letφbe the composite projection τ p φ 1 : Z 6 P 19 and V σ the subscheme of Z 6 corresponding to 6-tuples with a coincidence of type σ (see 4). A diagram of the situation is: X 6 j 6 X Z 6 j f φ1 Z p S τ P 19 The result giving the degree of some strata of cubic surfaces with a conic node is the following Theorem 5.1. Let σ =(k 1,..., k r ) be a partition of 6 with r 3 and let Λ be a linear space of P 19 of dimension 7 r. If Π=φ 1 (Λ), then the degree of the stratum C σ of cubic surfaces with a conic node of type σ is deg(c σ )= 1 [X 6 ] [V σ ] [Π], n(σ) Z 6 except for the case σ =(2, 2, 2) where deg(c σ )= 1 [X 6 ] [V σ ] [Π] 1 n(σ) Z 6 4 deg(c U ). In both formulas, n(σ) is an integer determined by the partition σ. Proof. We divide the proof into 4 steps. 1. We first translate the computation of deg(c σ )toa (S) showing that, if S σ S is the closure of the locus of cubic surfaces with a distinguished singular point such that the associated collection of lines through this point have

108 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO collapsed in accordance with the partition σ (i.e. τ(s σ )=C σ ), then if σ (2, 2, 2), m deg(c σ )= [S σ Λ S ], S where m is the number of conic nodes of a cubic surface in the stratum C σ and Λ S = τ 1 (Λ) (see Lemma 5.2). When σ =(2, 2, 2), τ(s σ )=C σ C U and therefore 4 deg(c σ )+deg(c U )= [S σ Λ S ]. 2. We define a subscheme W Z 6 by W = X 6 V σ Π (in this way (p φ 1 )(W )= S σ Λ S ) and we prove that [W ] A (Z 6 ) enumerates the ordered 6-tuples of φ 1 (W )overs if and only if dim(λ S ) 3, which is equivalent to r 3above (see Lemma 5.3). For the rest of the proof we then suppose that dim(λ S ) 3. 3. We show in Lemma 5.4 that the cycle [W ] A (Z 6 ) pushes forward to ñ(σ) [S σ Λ S ]ina (S), and we compute ñ(σ) according to the partition σ. 4. We verify in Lemma 5.5 that [W ]=[X 6 V σ Π] = [X 6 ] [V σ ] [Π]. In this way, the theorem is proved with n(σ) =m ñ(σ) depending only on the partition σ. Lemma 5.2. With the previous notation, let m be the number of conic nodes presented by a surface belonging to the stratum C σ,thenifσ (2, 2, 2), m deg(c σ )= [S σ Λ S ] and if σ =(2, 2, 2), m deg(c σ )+deg(c U )= [S σ Λ S ]. Proof. Suppose that σ (2, 2, 2): if the type of singularities presented by a cubic surface corresponding to a point of S σ is ma 1 A q (see 2) then deg(τ Sσ )=m (for the projection τ forgets the double point) and consequently τ ([S σ ]) = m [C σ ]. By construction, µ := c 1 (O S (1)) = τ H,whereH is the hyperplane class of P 19. Applying the projection formula, we then have m deg(c σ )=m [C σ ] H ασ = [S σ ] µ ασ = [S σ Λ S ]. P 19 S S The last equality is [S σ Λ S ]=[S σ ] [Λ S ]. This is true because the intersection S σ Λ S consists of points and it is transversal at each one of them. (If this last claim were false, it would be also false for τ(s σ Λ S )=C σ Λandweknowthat a general linear space in a projective space has a non-empty intersection with any subvariety of complementary dimension and one can choose it so the intersection is transversal by Bertini s theorem (see II.8.18 of [12])). If σ =(2, 2, 2), a surface corresponding to a point of S σ can have 4 conic nodes or else a uninode, so this time τ ([S σ ]) = 4 [C σ ]+[C U ] and the rest is as before. With the notation given in 4, if p : Z S, a cycle [A] Z k enumerates all the k-tuples of φ 1 (A) Z with the same image in S if and only if p φ1(a) is curvilinear. We recall that a map p : Z S is called curvilinear when dim(ω 1 p(z)) 1, z Z S S S

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 109 (with Ω 1 p the relative differentials (see B.2.7 of [9])) and that, if dim(z) = n, dim(s) =m and S i (p) :={z Z : rango(dp) j} = {z Z : dim(ω 1 p (z)) n j}, where j = min(n, m) i, is the locus of Thom-Boardman singularities of p (see [16]), then if n m, p curvilinear S 2 (p) =, andifn m, p curvilinear implies that S 2 (p) =. Lemma 5.3. The cycle [W ]=[X 6 V σ Π] enumerates all the (ordered) 6-tuples of φ 1 (W ) with the same image in S if and only if dim(λ S ) 3. Proof. Taking suitable scheme-theoretic inverse images of X through the inductive construction of X 6 and then going down to X by appropriated projections, one can define a subscheme X σ X parametrizing the points x X such that x 1,x 2,..., x 5 X with f(x) =f(x 1 )= = f(x 5 ), and the first k 1 points (included x) are i.n., the next k 2 points are i.n., and the last k r points are i.n., where i.n. means infinitely near points (see 2.3. of [7]). Recalling the previous definitions, we have φ 1 (W )=j(x σ f 1 Λ S ). Since j is a regular inclusion and f = p j, thenp φ1(w ) will be curvilinear when f Xσ f 1 Λ S is curvilinear. Now, f curvilinear S 2 (f) =, fordim(x) =dim(s). We then study the Thom-Boardman singularities locus of f, } S 2 (f) = {x X :dim(ω 1 f (x)) 2 = {x X : rk(df x ) dim(x) 2}, with df x : T x X T f(x) S, the map induced by f on the tangent spaces. By definition of S 2 (f), we have codim(s 2 (f),x) 2 2 = 4 (see (III, 5) of [16]). Checking that, in fact, for our f this codimension is 4, the proof is finished because in that case, when dim(λ S ) 3, f 1 (Λ S ) can be chosen so that f 1 (Λ S ) S 2 (f) = and it is easy to see that dim(λ S ) 3 r 3. Let us verify that codim(s 2 (f),x) = 4. We recall from (2.1) that, fixed P P 3, a cubic surface of P 3 with P singular, has equation x 3 f 2 + f 3 = 0, up to change of coordinates. The plane where the conic and cubic curves associated to f 2 and f 3, respectively, live is fixed and all possible surfaces correspond to all possible pairs of conic and cubic. If R = O P 2(2) O P 2(3) (see Proposition 3.11), then one gets X P P (H 0 (R)) P 2 = P 15 P 2 intersecting every pair of conic and cubic curves and, consequently, the part of T x X living in T w P 2, which eventually will contribute to the locus S 2 (f) is the intersection of the Zariski tangent spaces of the two curves at the point w. In that way, x S 2 (f) when this intersection has dimension 2 and so codim(s 2 (f),x) is the number of conditions for a pair of plane curves of degree 2 and 3 having an intersection of their tangent spaces at a common point of dimension 2. It is easy to see that this number is 4 (because that happens if and only if the two curves are singular at the common point). Lemma 5.4. With the previous notation: [W ]= ñ(σ) [S σ Λ S ], Z 6 S

110 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO where ñ(σ) is an integer determined by the partition σ according to the following table: σ ñ(σ) (1, 1, 1, 1, 1, 1) 6! = 5! 6 (2, 1, 1, 1, 1) 24 = 4! 1 (3, 1, 1, 1) 6=3! 1 (2, 2, 1, 1) 4=2 2 (4, 1, 1) 2=2 1 (3, 2, 1) 1 (2, 2, 2) 6=2 3 Proof. By construction, p φ1(w )= S σ Λ S,so [W ]= deg((p φ 1 ) W )[S σ Λ S ], Z 6 S and the factor ñ(σ) isdeg((p φ 1 ) W )=deg(φ 1 W ) deg(p φ1(w )). When p : Z S is curvilinear, there is a natural closed inclusion Z k Z k = Z S S Z, which is an isomorphism outside the diagonals. Each permutation between the factors of Z k induces an automorphism of Z k (see 4.7.2. of [18]). In our case, when σ =(k 1,..., k r )andr 3, then p φ1(w ) is curvilinear; W is included in the intersection σ = 1,...,k1 k1+1,...,k 1+k 2 k1+...+k r 1+1,...,6 Z 6 (see 3.3 of [7]) and, this time, the automorphisms of W are the permutations interchanging diagonals of the same size. Since the image by φ 1 of a 6-tuple in W is its first element, then: -deg(φ 1 W ) is the number of automorphisms of W fixing the first coordinate (see 3.3 of [7]), - if the first element on the partition σ =(k 1,k 2,..., k r ) is repeated l times, then deg(p φ1(w )) =l. Lemma 5.5. With the previous notation, [X 6 V σ Π] = [X 6 ] [V σ ] [Π]. Z 6 Z 6 Proof. The cycle [X 6 V σ Π] is zero-dimensional (we have seen that [X 6 V σ Π] = ñ(σ) [S σ Λ S ] Z 6 S and that [S σ Λ S ] is zero-dimensional). We recall from 8.1.11 of [9] that if U and V are two non-singular varieties, it is enough that the intersection is transversal at generic points in order to have [U V ]=[U] [V]. In our case, this result applies because: -Π=φ 1 (Λ) is the inverse image by the smooth morphism φ of a linear space Λ P 19 and therefore Π intersects every component of X 6 V σ. - The full intersection is tranversal at each point, because their image by φ in P 19 is tranversal. - X 6 V σ is also tranversal since, if it were not, X 6 V σ Π and its image in P 19 will be non-reduced and this is not the case.

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 111 Now, the proof of Theorem 5.1 is complete and the problem of getting the degree of a stratum C σ consists in determining [X 6 ] and computing [X 6 ] [V σ ] [Π] in the ring A (Z 6 ). We are going to get [X 6 ] using a result by Z. Ran [24] based on the formula of residual intersection of W. Fulton and D. Laksov [10]. Proposition 5.6. With the previous notation, ( { } ) [X 6 ]=Π 6 j=1 (p j j,j ) ([X]) [D j,6 (p ] j,j ) c(e) 1+[D j,6 ] 1 where p j j,j = p2 2,2 p 3 2,3 p j 2,j, Dj,6 = D1,j 6 + D6 2,j + + D6 j 1,j and D6 i,j = (p 6 i,j ) D j i+1 1,j i+1 with p6 i,j a composition of projections from Z 6 to Z j i+1 onto the first factor if i =1and onto the second factor if i>1. Proof. To get [X 6 ] we can apply Theorem (4.2) from [24] because: -sincex =(s E ) 0 where s E is a section of the vector bundle E over Z, then c(n X Z)=j c(e). - the morphisms X t X t 1 induced by f : X S are local complete intersection (l.c.i.) of the same codimension, for t = 1,..., k, so the residual intersection formula of [10] applies. To see this, we first note that f is l.c.i. of codimension 0 because X Z is a closed regular inclusion of codimension 2 and Z is smooth over S with relative dimension 2. Now, in general, if X is Cohen-Macaulay and f is l.c.i. of codimension n, then a necessary and sufficient condition for every morphism X t X t 1 with t =1,..., k, to be l.c.i. of codimension n is that all of these morphisms are of the same codimension as f (see [17]). As we saw in 3, X is smooth and X is also the zero scheme of a regular section s E of the vector bundle E over Z. In this case, by the inductive constructions of 4, each X t is again smooth and it is also the zero locus of a regular section of the secant bundle E t over Z t,socodim(x t,z t )=rank(e t )(forz t is Cohen-Macaulay (in fact, smooth)) and, finally X t X t 1 is l.c.i. of codimension 0 for all t. The formula of residual intersection ((4.2) of [24]) says that ( { (p j [X k ]=Π k j=1 (p j j,j ) ([X]) [D j,k ] j,j ) c(e) 1+[D j,k ] where { } r denotes the homogeneous component of dim r and c(e) and[x] =c 2 (E) are known (see Lemma 3.10). Lemma 5.7. If V σ is the intersection of (the pullback of) the exceptional divisors E 1,..., E j,then[v σ ]=[E 1 ] [E j ]. Proof. Using 8.1.11 of [9], as in Lemma 5.5, it is enough to note that the intersection of exceptional divisors V σ = j l=1 E l is not only transversal but it is smooth. 6. Effective computation of degree and intersection ring of spaces for k-tuples In order to determine deg(c σ )wehavetocompute[x 6 ] [V σ ] [Π] in A (Z 6 ). In general, if Z is a projective bundle over S and A (S) is known, we can determine completely A (Z k ) for any k =1, 2,... (see Proposition 6.2) and, in our case, it turns out to be a polynomial ring on 23 variables, modulo 38 relations, so effective computations in this ring, using Gröbner bases methods are very costly. } 2 1 ),

112 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO In spite of this fact, in order to effectively compute the product [X 6 ] [V σ ] [Π] in A (Z 6 )itisnotnecessarytoknowthewholering,aswewillseeinamoment. In this section the notation is the same as in 4, writing ec k,1 := [D k 1,k ]forthe class of the exceptional divisor in Z k. We will apply several times a result by S. Keel [15] about the intersection ring of a blown-up variety, saying that, if i : X Y is a regular inclusion of codimension d, i : A (Y ) A (X) is surjective and Ỹ Y is the blow up of Y along X with exceptional divisor X, then A (Ỹ ) A (Y )[T ] (6.1) = (P (T ),T ker(i )) where P (T ) A (Y )[T ] is any polynomial with constant term equal to [X] and whose restriction to A (X) is the Chern polynomial of the normal bundle N X Y. Proposition 6.1. Let p : Z S be a smooth morphism and suppose that A (Z) is a finitely generated A (S)-module. For k =1, 2,... let a A (Z k ) be a zerodimensional cycle consisting in a product of pullbacks of cycles from A (Z j ) with j k and some power of the class of the exceptional divisor in Z k.then,thedegree of the cycle a can be computed as the degree of a zero-dimensional cycle in A (Z). Proof. We prove this by induction in k. The cases k =1, 2 are easy to work out, so let s suppose that we can determine the degree of a zero-dimensional cycle in A (Z k ) going down to A (Z k 1 ) and let s prove that we can do the same with one in A (Z k+1 ). By Keel s result (6.1) described before, a = ec α k+1,1 b k+1 c with c A (Z k Zk 1 Z k ). We distinguish two cases: 1. α = 0. In this case, as b k+1 b k+1c = c (see 6.7.(b) of [9]), we have a = b k+1c = c = pr 1 c. Z k+1 Z k+1 Z k Zk 1 Z k Z k To get pr 1 c, we recall the diagram E k+1 j k+1 Z k+1 g k+1 b k+1 k l k pr 2 Zk Zk 1 Z k pr1 Zk ι k 1 p k 1,k 1 p k 2,k Z k Z k 1 Since, by hypothesis c A (Z k Zk 1 Z k ) can be written as c = pr1 d pr 2 e with d, e A (Z k ) then, denoting by q 1 := ι k 1 p k 1,k 1 and q 2 := p k 2,k and using the projection formula, we have pr 1 c = d pr 1 pr2 e = d q2 q 1 e, Z k Z k Z k where the last equality is (1.7) of [9]. To know the product d q2q 1 e A (Z k ) we need to determine q 1 e.now, by Keel s result (6.1) again, e A (Z k )mustbee = ec γ k,1 b ku for some γ, with u A (Z k 1 Zk 2 Z k 1 ). Then: (a) If γ>0, q 1 e = ι k 1 pr 1 b k (ec γ k,1 b k u)=ι k 1 (pr 1 (u b k (ec γ k,1 ))) by projection formula. Writing U := pr 1 (u b k (ec γ k,1 )), we have that U =( 1) γ 1 m γ 2 lk 1 u A (Z k 1 ) = A ( k 1 ), being m γ 2 the term of degree γ 2enc(N k 1 ) 1 (see 4.2.2. of [9]).

DEGREE OF STRATA OF SINGULAR CUBIC SURFACES 113 It is easy to get U such that U = ι k 1 U,andthen:q 1 e = U so a = Z k+1 c = Z k Zk 1 Z k d q2u, Z k which is known by induction. (b) If γ = 0, q 1 e = ι k 1 pr 1 b k b k u = ι k 1 pr 1 u. Since u A (Z k 1 Zk 2 Z k 1 ), we have again, by hypothesis, that u = pr1 f pr 2 h with f,h A (Z k 1 ) and, consequently pr 1 u = f pr 1 pr2h. Calling now V := pr 1 u, and working as in (a), we find a = Z k+1 c = Z k Zk 1 Z k d q2 V, Z k which is known again by induction. 2. If a = ec α k+1,1 b k+1c with α>0, then the situation is similar to (a) when we computed U, so using 4.2.2. of [9], we conclude a = Z k+1 pr 1 b k+1 a = Z k ( 1) α 1 m α 2 lk c Z k (where again m α 2 is the term of degree α 2inc(N k ) 1 ), and this is known by induction. In order to effectively apply Proposition 6.1, we need to determine c(n k )for every k. From 4, we know that N k is isomorphic to the relative tangent bundle T k := T p k 2,k. The total Chern class of T k can be computed inductively, starting with c(t 1 ) the class of a relative tangent bundle (given by B.5.8. of [9]). Using standard results about Chern classes (see 15.4.(iv), 15.4.2. of [9]), one gets c(t k )=(1+ec k,1 ) 2 (1 ec k,1 ) i (p k 1,k 1 ) c 2 i (T k 1 ). i=0 Implementing the inductive method given in Proposition 6.1 in Maple [20] we finally get the degree of the strata of singular cubic surfaces with a conic node mentioned in Theorem 5.1: σ 1 6 21 4 31 3 2 2 1 2 41 2 321 2 3 Singularity A 1 2A 1 A 1 A 2 3A 1 A 1 A 3 2A 1 A 2 4A 1 Parameters 3 2 1 1 0 0 0 deg(c σ ) 32 280 1200 800 2200 3420 305 Note that for σ =(2, 2, 2), we have substracted the contribution of surfaces with a uninode which is 1 4 deg(c U), according to Theorem 5.1. The degree of the stratum C U, 260, can be computed separately at least in two ways. First, as a particular case of some formulas for the degree of varieties parametrizing quadrics with vertex of specified dimension, given by I. Vainsencher [28] (1979). Secondly, as we do in [30], using a result by J. Harris and L. W. Tu [11] (1984) about the cohomology class of the degeneracy locus of a symmetric bundle map. For completeness, we describe here the intersection ring of the space Z k of k- tuples of Z with the same image in S (defined in 4), when Z = P (E) π S is the projectivization of a vector bundle E over a smooth scheme S.

114 RAFAEL HERNÁNDEZ AND MARÍA J. VÁZQUEZ-GALLO Proposition 6.2. Let Z be a projective bundle over a smooth scheme S and let Z k be the space of k-tuples of Z over S defined in 4. Then A (Z k ) = A (S)[g 1,..., g m ], (r 1,..., r n ) where m = k j=1 j, n = k j=1 (2j 1); k of the generators g j are (pullback of) the hyperplane class of Z; the rest are (pullback of) exceptional divisors classes and the relations r j are determined by 3.3.(b), Ex. 8.3.4. of [9] and Keel s result (6.1) above. Proof. We see this by induction on k, using repeatedly 3.3.(b), Ex. 8.3.4. of [9], to determine the ring of a projective bundle and Keel s work (6.1) from [15], for the ring of a blown-up scheme. One starts with the projective bundle Z and its Chow ring A (Z) written in terms of A (S). For k = 2, by construction, the fiber product Z S Z is another projective bundle, one finds that A (Z S Z) = A (Z) A (S)A (Z), and Z 2 = Bl (Z S Z) is a blown-up projective bundle. When k =3checkingthatZ 2 Z Z 2 is a projective bundle over Z 2 (by 3.4. of [13]) one has that Z 3 is a projective bundle blown-up two times. At the step (k 1), one supposes that Z k 1 is a projective bundle over Z k 2 blown-up (k 1) times and therefore its Chow ring can be written as it is said in the proposition. Now, again by definition, the fiber product Z k 1 Zk 2 Z k 1 is another projective bundle over Z k 1 blown-up (k 1) times and, finally, for Z k one adds another blow-up getting A (Z k ) as the ring described in the proposition (for details see [30]). References 1. A. B. Altman and S. L. Kleiman, Foundations of the theory of Fano schemes, Compos.Math. 34 (1977), 3-47. MR 58:27967 2. V. I. Arnold, Normal forms of functions near degenerate critical points, the Weyl groups A k, D k and E k and Lagrangian singularities, Func. Anal. Appl. 6 (1972), 254-272. MR 50:8595 3. V. I. Arnold, Normal forms of functions in the neighborhood of degenerate critical points, Russ. Math. Surv. 29 (2) (1974), 11-49. MR 58:24324 4. J. W. Bruce, A stratification of the space of cubic surfaces, Math. Proc. Cambridge 87 (1980), 427-441. MR 81c:14022 5. J. W. Bruce and C. T. C. Wall, On the clasification of cubic surfaces, J.Lond.Math.Soc. (2) 19 (1979), 245-256. MR 80f:14021 6. A. Cayley, A memoir on cubic surfaces, Phil. Trans. Roy. Soc. 159 (1869), 231-236. 7. S. J. Colley, Enumerating stationary multiple-points, Adv. Math. 66 (1987), 149-170. MR 89g:14042 8. D. F. Coray and I. Vainsencher, Enumerative formulae for ruled cubic surfaces and rational quintic curves, Comment. Math. Helv. 61 (1986), 510-518. MR 87m:14061 9. W. Fulton, Intersection Theory, Springer-Verlag, (1984). MR 85k:14004 10. W. Fulton and D. Laksov, Residual intersections and the double point formula, Real and complex singularities, ed. P. Holm, Sijthoff & Nordhoff, (1977), 171-177. MR 58:27971 11. J. Harris and L. W. Tu, On symmetric and skew-symmetric determinantal varieties, Topology, Vol. 23, No.1 (1984), 71-84. MR 85c:14032 12. R. Hartshorne, Algebraic Geometry, Springer-Verlag, (1977). MR 57:3116 13. S. Iitaka, Algebraic Geometry, Springer Verlag, (1982). MR 84j:14001 14. S. Katz and S. A. Stromme, Schubert: a Maple package for intersection theory. Available by anonymous ftp from ftp.math.okstate.edu or linus.mi.uib.no, cd pub/schubert. 15. S. Keel, Intersection theory of moduli space of stable N-pointed curves of genus cero, Trans. Amer. Math. Soc. 330 (1992), 545-574. MR 92f:14003 16. S. L. Kleiman, The enumerative theory of singularities, Real and complex singularities, ed. P. Holm, Sijthoff & Nordhoff, (1977), 297-396. MR 58:27960